It looks like metapage_releasepage was making in invalid assumption that
the releasepage method would not be called on a dirty page. Instead of
issuing a warning and releasing the metapage, it should return 0, indicating
that the private data for the page cannot be released.
I also realized that metapage_releasepage had the return code all wrong. If
it is successful in releasing the private data, it should return 1, otherwise
it needs to return 0.
Lastly, there is no need to call wait_on_page_writeback, since
try_to_release_page will not call us with a page in writback state.
Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Else a subsequent bio_clone might make a mess.
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: "Don Dupuis" <dondster@gmail.com>
Acked-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Both cause the 'entries' count in the export cache to be non-zero at module
removal time, so unregistering that cache fails and results in an oops.
1/ exp_pseudoroot (used for NFSv4 only) leaks a reference to an export
entry.
2/ sunrpc_cache_update doesn't increment the entries count when it adds
an entry.
Thanks to "david m. richter" <richterd@citi.umich.edu> for triggering the
problem and finding one of the bugs.
Cc: "david m. richter" <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Don't reassign to watch. If idr_find() returns NULL, then
put_inotify_watch() will choke.
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
While doing some inotify stress testing, I hit the following race. In
inotify_release(), it's possible for a watch to be removed from the lists
in between dropping dev->mutex and taking inode->inotify_mutex. The
reference we hold prevents the watch from being freed, but not from being
removed.
Checking the dev's idr mapping will prevent a double list_del of the
same watch.
Signed-off-by: Amy Griffis <amy.griffis@hp.com>
Acked-by: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rml@novell.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Bernd Schmidt points out that binfmt_flat is now leaving the exec file open
while the application runs. This offsets all the application's fd numbers.
We should have closed the file within exec(), not at exit()-time.
But there doesn't seem to be a lot of point in doing all this just to avoid
going over RLIMIT_NOFILE by one fd for a few microseconds. So take the EMFILE
checking out again. This will cause binfmt_flat to again fail LTP's
exec-should-return-EMFILE-when-fdtable-is-full test. That test appears to be
wrong anyway - Open Group specs say nothing about exec() returning EMFILE.
Cc: Bernd Schmidt <bernd.schmidt@analog.com>
Cc: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Assigning the result of posix_acl_to_xattr() to an unsigned data type
(size/size_t) obscures possible errors.
Coverity CID: 1206.
Signed-off-by: Florin Malita <fmalita@gmail.com>
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Address a problem found when a Linux NFS server uses the "subtree_check"
export option.
The "subtree_check" NFS export option was designed to prohibit a client
from using a file handle for which it should not have permission. The
algorithm used is to ensure that the entire path to the file being
referenced is accessible to the user attempting to use the file handle. If
some part of the path is not accessible, then the operation is aborted and
the appropriate version of ESTALE is returned to the NFS client.
The error, ESTALE, is unfortunate in that it causes NFS clients to make
certain assumptions about the continued existence of the file. They assume
that the file no longer exists and refuse to attempt to access it again.
In this case, the file really does exist, but access was denied by the
server for a particular user.
A better error to return would be an EACCES sort of error. This would
inform the client that the particular operation that it was attempting was
not allowed, without the nasty side effects of the ESTALE error.
Signed-off-by: Peter Staubach <staubach@redhat.com>
Acked-By: NeilBrown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Functions compat_nfs_svc_trans, compat_nfs_clnt_trans,
compat_nfs_exp_trans, compat_nfs_getfd_trans and compat_nfs_getfs_trans,
which are called by compat_sys_nfsservctl(fs/compat.c), don't handle the
return value of access_ok properly. access_ok return 1 when the addr is
valid, and 0 when it's not, but these functions have the reversed
understanding. When the address is valid, they always return -EFAULT to
compat_sys_nfsservctl.
An example is to run /usr/sbin/rpc.nfsd(32bit program on Power5). It
doesn't function as expected. strace showes that nfsservctl returns
-EFAULT.
The patch fixes this by correcting the error handling on the return value
of access_ok in the five functions.
Signed-off-by: Lin Feng Shen <shenlinf@cn.ibm.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
configfs_init() needs to be called first to register configfs before anyconsumers try to access it. Move up configfs in fs/Makefile to make
sure it is initialized early.
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
If configfs_mkdir() errored in certain ways after the parent<->child
linkage was already created, it would not undo the linkage. Also,
comment the reference counting for clarity.
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
configfs_mkdir() failed to release the working parent reference in most
exit paths. Also changed the exit path for readability.
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
We were using GFP_KERNEL in a handful of places which really wanted
GFP_NOFS. Fix this.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Temporarily take the meta data lock in ocfs2_file_aio_read() to allow us to
update our inode fields.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
We need to take a data lock around extends to protect the pages that
ocfs2_zero_extend is going to be pulling into the page cache. Otherwise an
extend on one node might populate the page cache with data pages that have
no lock coverage.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
fs/jffs2/nodelist.c: In function `check_node_data':
fs/jffs2/nodelist.c:441: warning: unsigned int format, different type arg (arg 4)
fs/jffs2/nodelist.c:464: warning: int format, different type arg (arg 5)
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Revert commit f6422f17d3, due to
Valdis.Kletnieks@vt.edu wrote:
>
> There seems to have been a bug introduced in this changeset:
>
> Am running 2.6.17-rc3-mm1. When this changeset is applied, 'mount --bind'
> misbehaves:
>
> > # mkdir /foo
> > # mount -t tmpfs -o rw,nosuid,nodev,noexec,noatime,nodiratime none /foo
> > # mkdir /foo/bar
> > # mount --bind /foo/bar /foo
> > # tail -2 /proc/mounts
> > none /foo tmpfs rw,nosuid,nodev,noexec,noatime,nodiratime 0 0
> > none /foo tmpfs rw 0 0
>
> Reverting this changeset causes both mounts to have the same options.
>
> (Thanks to Stephen Smalley for tracking down the changeset...)
>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: <Valdis.Kletnieks@vt.edu>
Cc: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Multiple races can happen when v9fs is interrupted by a signal and Tflush
message is sent to the server. After v9fs sends Tflush it doesn't wait
until it receives Rflush, and possibly the response of the original
message. This behavior may confuse v9fs what fids are allocated by the
file server.
This patch fixes the races and the fid allocation.
Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@hera.kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
v9fs leaks memory if the file server responds with Rerror to a Twalk
message. The patch fixes the leak.
Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@hera.kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Yesterday, I got the following error with 2.6.16.13 during a file copy from
a smb filesystem over a wireless link. I guess there was some error on the
wireless link, which in turn caused an error condition for the smb
filesystem.
In the log, smb_file_read reports error=4294966784 (0xfffffe00), which also
shows up in the slab dumps, and also is -ERESTARTSYS. Error code 27499
corresponds to 0x6b6b, so the rq_errno field seems to be the only one being
set after freeing the slab.
In smb_add_request (which is the only place in smbfs where I found
ERESTARTSYS), I found the following:
if (!timeleft || signal_pending(current)) {
/*
* On timeout or on interrupt we want to try and remove the
* request from the recvq/xmitq.
*/
smb_lock_server(server);
if (!(req->rq_flags & SMB_REQ_RECEIVED)) {
list_del_init(&req->rq_queue);
smb_rput(req);
}
smb_unlock_server(server);
}
[...]
if (signal_pending(current))
req->rq_errno = -ERESTARTSYS;
I guess that some codepath like smbiod_flush() caused the request to be
removed from the queue, and smb_rput(req) be called, without
SMB_REQ_RECEIVED being set. This violates an asumption made by the quoted
code.
Then, the above code calls smb_rput(req) again, the req gets freed, and
req->rq_errno = -ERESTARTSYS writes into the already freed slab. As
list_del_init doesn't cause an error if called multiple times, that does
cause the observed behaviour (freed slab with rq_errno=-ERESTARTSYS).
If this observation is correct, the following patch should fix it.
I wonder why the smb code uses list_del_init everywhere - using list_del
instead would catch such situations by poisoning the next and prev
pointers.
May 4 23:29:21 knautsch kernel: [17180085.456000] ipw2200: Firmware error detected. Restarting.
May 4 23:29:21 knautsch kernel: [17180085.456000] ipw2200: Sysfs 'error' log captured.
May 4 23:33:02 knautsch kernel: [17180306.316000] ipw2200: Firmware error detected. Restarting.
May 4 23:33:02 knautsch kernel: [17180306.316000] ipw2200: Sysfs 'error' log already exists.
May 4 23:33:02 knautsch kernel: [17180306.968000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:34:18 knautsch kernel: [17180383.256000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:34:18 knautsch kernel: [17180383.284000] SMB connection re-established (-5)
May 4 23:37:19 knautsch kernel: [17180563.956000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:40:09 knautsch kernel: [17180733.636000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:40:26 knautsch kernel: [17180750.700000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:43:02 knautsch kernel: [17180907.304000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:43:08 knautsch kernel: [17180912.324000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:43:34 knautsch kernel: [17180938.416000] smb_errno: class Unknown, code 27499 from command 0x6b
May 4 23:43:34 knautsch kernel: [17180938.416000] Slab corruption: start=c4ebe09c, len=244
May 4 23:43:34 knautsch kernel: [17180938.416000] Redzone: 0x5a2cf071/0x5a2cf071.
May 4 23:43:34 knautsch kernel: [17180938.416000] Last user: [<e087b903>](smb_rput+0x53/0x90 [smbfs])
May 4 23:43:34 knautsch kernel: [17180938.416000] 000: 6b 6b 6b 6b 6b 6b 6b 6b 6a 6b 6b 6b 6b 6b 6b 6b
May 4 23:43:34 knautsch kernel: [17180938.416000] 0f0: 00 fe ff ff
May 4 23:43:34 knautsch kernel: [17180938.416000] Next obj: start=c4ebe19c, len=244
May 4 23:43:34 knautsch kernel: [17180938.416000] Redzone: 0x5a2cf071/0x5a2cf071.
May 4 23:43:34 knautsch kernel: [17180938.416000] Last user: [<00000000>](_stext+0x3feffde0/0x30)
May 4 23:43:34 knautsch kernel: [17180938.416000] 000: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
May 4 23:43:34 knautsch kernel: [17180938.416000] 010: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
May 4 23:43:34 knautsch kernel: [17180938.460000] SMB connection re-established (-5)
May 4 23:43:42 knautsch kernel: [17180946.292000] ipw2200: Firmware error detected. Restarting.
May 4 23:43:42 knautsch kernel: [17180946.292000] ipw2200: Sysfs 'error' log already exists.
May 4 23:45:04 knautsch kernel: [17181028.752000] ipw2200: Firmware error detected. Restarting.
May 4 23:45:04 knautsch kernel: [17181028.752000] ipw2200: Sysfs 'error' log already exists.
May 4 23:45:05 knautsch kernel: [17181029.868000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:45:36 knautsch kernel: [17181060.984000] smb_errno: class Unknown, code 27499 from command 0x6b
May 4 23:45:36 knautsch kernel: [17181060.984000] Slab corruption: start=c4ebe09c, len=244
May 4 23:45:36 knautsch kernel: [17181060.984000] Redzone: 0x5a2cf071/0x5a2cf071.
May 4 23:45:36 knautsch kernel: [17181060.984000] Last user: [<e087b903>](smb_rput+0x53/0x90 [smbfs])
May 4 23:45:36 knautsch kernel: [17181060.984000] 000: 6b 6b 6b 6b 6b 6b 6b 6b 6a 6b 6b 6b 6b 6b 6b 6b
May 4 23:45:36 knautsch kernel: [17181060.984000] 0f0: 00 fe ff ff
May 4 23:45:36 knautsch kernel: [17181060.984000] Next obj: start=c4ebe19c, len=244
May 4 23:45:36 knautsch kernel: [17181060.984000] Redzone: 0x5a2cf071/0x5a2cf071.
May 4 23:45:36 knautsch kernel: [17181060.984000] Last user: [<00000000>](_stext+0x3feffde0/0x30)
May 4 23:45:36 knautsch kernel: [17181060.984000] 000: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
May 4 23:45:36 knautsch kernel: [17181060.984000] 010: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
May 4 23:45:36 knautsch kernel: [17181061.024000] SMB connection re-established (-5)
May 4 23:46:17 knautsch kernel: [17181102.132000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:47:46 knautsch kernel: [17181190.468000] smb_errno: class Unknown, code 27499 from command 0x6b
May 4 23:47:46 knautsch kernel: [17181190.468000] Slab corruption: start=c4ebe09c, len=244
May 4 23:47:46 knautsch kernel: [17181190.468000] Redzone: 0x5a2cf071/0x5a2cf071.
May 4 23:47:46 knautsch kernel: [17181190.468000] Last user: [<e087b903>](smb_rput+0x53/0x90 [smbfs])
May 4 23:47:46 knautsch kernel: [17181190.468000] 000: 6b 6b 6b 6b 6b 6b 6b 6b 6a 6b 6b 6b 6b 6b 6b 6b
May 4 23:47:46 knautsch kernel: [17181190.468000] 0f0: 00 fe ff ff
May 4 23:47:46 knautsch kernel: [17181190.468000] Next obj: start=c4ebe19c, len=244
May 4 23:47:46 knautsch kernel: [17181190.468000] Redzone: 0x5a2cf071/0x5a2cf071.
May 4 23:47:46 knautsch kernel: [17181190.468000] Last user: [<00000000>](_stext+0x3feffde0/0x30)
May 4 23:47:46 knautsch kernel: [17181190.468000] 000: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
May 4 23:47:46 knautsch kernel: [17181190.468000] 010: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b
May 4 23:47:46 knautsch kernel: [17181190.492000] SMB connection re-established (-5)
May 4 23:49:20 knautsch kernel: [17181284.828000] smb_file_read: //some_file validation failed, error=4294966784
May 4 23:49:39 knautsch kernel: [17181303.896000] smb_file_read: //some_file validation failed, error=4294966784
Signed-off-by: Jan Niehusmann <jan@gondor.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Mark Moseley reported that a chroot environment on a SMB share can be left
via "cd ..\\". Similar to CVE-2006-1863 issue with cifs, this fix is for
smbfs.
Steven French <sfrench@us.ibm.com> wrote:
Looks fine to me. This should catch the slash on lookup or equivalent,
which will be all obvious paths of interest.
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch fixes two problems.
First, the comparison of entries in the waitq.c was incorrect.
Second, the NFY_NONE check was incorrect. The test of whether the dentry
is mounted if ineffective, for example, if an expire fails then we could
wait forever on a non existant expire. The bug was identified by Jeff
Moyer.
The patch changes autofs4 to wait on expires only as this is all that's
needed. If there is no existing wait when autofs4_wait is call with a type
of NFY_NONE it delays until either a wait appears or the the expire flag is
cleared.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Make sure to clear the driverfs_dev pointer when we do del_gendisk() (on
disk removal), so that other users that may still have a ref to the disk
won't try to use the stale pointer.
Also move the KOBJ_REMOVE uevent handler up, so that the uevent still
has access to the driverfs_dev data.
This all should hopefully fix the problems with MMC umounts after device
removals that caused commit 56cf6504fc and
its reversal (1a2acc9e92).
Original problem reported by Todd Blumer and others.
Acked-by: Greg KH <gregkh@suse.de>
Cc: Russell King <rmk+lkml@arm.linux.org.uk>
Cc: James Bottomley <James.Bottomley@SteelEye.com>
Cc: Erik Mouw <erik@harddisk-recovery.com>
Cc: Andrew Vasquez <andrew.vasquez@qlogic.com>
Cc: Todd Blumer <todd@sdgsystems.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
* git://oss.sgi.com:8090/xfs-2.6:
[XFS] Fix a possible metadata buffer (AGFL) refcount leak when fixing an
[XFS] Fix a project quota space accounting leak on rename.
[XFS] Fix a possible forced shutdown due to mishandling write barriers
It is insane to be giving lease_init() the task of freeing the lock it is
supposed to initialise, given that the lock is not guaranteed to be
allocated on the stack. This causes lockups in fcntl_setlease().
Problem diagnosed by Daniel Hokka Zakrisson <daniel@hozac.com>
Also fix a slab leak in __setlease() due to an uninitialised return value.
Problem diagnosed by Björn Steinbrink.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Tested-by: Daniel Hokka Zakrisson <daniel@hozac.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
nr_segs may not be > UIO_MAXIOV, however it may be equal to. This makes
the behaviour identical to the real sys_vmsplice(). The other foov
syscalls also agree that this is the way to go.
Signed-off-by: Jens Axboe <axboe@suse.de>
This can happen quite easily, if several processes are trying to splice
the same file at the same time. It's not a failure, it just means someone
raced with us in allocating this file page. So just dump the allocated
page and relookup the original.
Signed-off-by: Jens Axboe <axboe@suse.de>
Nick says that the current construct isn't safe. This goes back to the
original, but sets PIPE_BUF_FLAG_LRU on user pages as well as they all
seem to be on the LRU in the first place.
Signed-off-by: Jens Axboe <axboe@suse.de>
Looking at generic_file_buffered_write(), we need to unlock_page() if
prepare write fails and it isn't due to racing with truncate().
Also trim the size if ->prepare_write() fails, if we have to.
Signed-off-by: Jens Axboe <axboe@suse.de>
Some places in ext3 multiple block allocation code (in 2.6.17-rc3) don't
handle the little endian well. This was resulting in *wrong* block numbers
being assigned to in-memory block variables and then stored on disk
eventually. The following patch has been verified to fix an ext3
filesystem failure when run ltp test on a 64 bit machine.
Signed-off-by; Mingming Cao <cmm@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Currently we rely on the PIPE_BUF_FLAG_LRU flag being set correctly
to know whether we need to fiddle with page LRU state after stealing it,
however for some origins we just don't know if the page is on the LRU
list or not.
So remove PIPE_BUF_FLAG_LRU and do this check/add manually in pipe_to_file()
instead.
Signed-off-by: Jens Axboe <axboe@suse.de>
We need to use the minium of {len, PAGE_SIZE-off}, not {len, PAGE_SIZE}-off.
The latter doesn't make any sense, and could cause us to attempt negative
length transfers...
Signed-off-by: Jens Axboe <axboe@suse.de>
* 'splice' of git://brick.kernel.dk/data/git/linux-2.6-block:
[PATCH] vmsplice: allow user to pass in gift pages
[PATCH] pipe: enable atomic copying of pipe data to/from user space
[PATCH] splice: call handle_ra_miss() on failure to lookup page
[PATCH] Add ->splice_read/splice_write to def_blk_fops
[PATCH] pipe: introduce ->pin() buffer operation
[PATCH] splice: fix bugs in pipe_to_file()
[PATCH] splice: fix bugs with stealing regular pipe pages
If SPLICE_F_GIFT is set, the user is basically giving this pages away to
the kernel. That means we can steal them for eg page cache uses instead
of copying it.
The data must be properly page aligned and also a multiple of the page size
in length.
Signed-off-by: Jens Axboe <axboe@suse.de>
The pipe ->map() method uses kmap() to virtually map the pages, which
is both slow and has known scalability issues on SMP. This patch enables
atomic copying of pipe pages, by pre-faulting data and using kmap_atomic()
instead.
lmbench bw_pipe and lat_pipe measurements agree this is a Good Thing. Here
are results from that on a UP machine with highmem (1.5GiB of RAM), running
first a UP kernel, SMP kernel, and SMP kernel patched.
Vanilla-UP:
Pipe bandwidth: 1622.28 MB/sec
Pipe bandwidth: 1610.59 MB/sec
Pipe bandwidth: 1608.30 MB/sec
Pipe latency: 7.3275 microseconds
Pipe latency: 7.2995 microseconds
Pipe latency: 7.3097 microseconds
Vanilla-SMP:
Pipe bandwidth: 1382.19 MB/sec
Pipe bandwidth: 1317.27 MB/sec
Pipe bandwidth: 1355.61 MB/sec
Pipe latency: 9.6402 microseconds
Pipe latency: 9.6696 microseconds
Pipe latency: 9.6153 microseconds
Patched-SMP:
Pipe bandwidth: 1578.70 MB/sec
Pipe bandwidth: 1579.95 MB/sec
Pipe bandwidth: 1578.63 MB/sec
Pipe latency: 9.1654 microseconds
Pipe latency: 9.2266 microseconds
Pipe latency: 9.1527 microseconds
Signed-off-by: Jens Axboe <axboe@suse.de>
The ->map() function is really expensive on highmem machines right now,
since it has to use the slower kmap() instead of kmap_atomic(). Splice
rarely needs to access the virtual address of a page, so it's a waste
of time doing it.
Introduce ->pin() to take over the responsibility of making sure the
page data is valid. ->map() is then reduced to just kmap(). That way we
can also share a most of the pipe buffer ops between pipe.c and splice.c
Signed-off-by: Jens Axboe <axboe@suse.de>
Found by Oleg Nesterov <oleg@tv-sign.ru>, fixed by me.
- Only allow full pages to go to the page cache.
- Check page != buf->page instead of using PIPE_BUF_FLAG_STOLEN.
- Remember to clear 'stolen' if add_to_page_cache() fails.
And as a cleanup on that:
- Make the bottom fall-through logic a little less convoluted. Also make
the steal path hold an extra reference to the page, so we don't have
to differentiate between stolen and non-stolen at the end.
Signed-off-by: Jens Axboe <axboe@suse.de>
- Check that page has suitable count for stealing in the regular pipes.
- pipe_to_file() assumes that the page is locked on succesful steal, so
do that in the pipe steal hook
- Missing unlock_page() in add_to_page_cache() failure.
Signed-off-by: Jens Axboe <axboe@suse.de>
